vLLM AsyncLLM mode is already available in verl, whose performance is better than vLLM sync mode. Any plan to support AsyncLLM ? related code: https://github.com/volcengine/verl/blob/main/verl/workers/rollout/vllm_rollout/vllm_async_server.py#L181 https://github.com/volcengine/verl/commit/aacd3660fc6697dc0853ff6c54872a5ca3ae9715
vLLM AsyncLLM mode is already available in verl, whose performance is better than vLLM sync mode.
Any plan to support AsyncLLM ?
related code:
https://github.com/volcengine/verl/blob/main/verl/workers/rollout/vllm_rollout/vllm_async_server.py#L181
verl-project/verl@aacd366